You are on page 1of 6

2019 IEEE Jordan International Joint Conference on Electrical Engineering and Information Technology (JEEIT)

Comparison Between the Dynamic Programming


and Particle Swarm Optimization for Solving Unit
Commitment Problems
Venkata Silpa Borra K. Debnath
Charles Darwin University Charles Darwin University
Darwin, Australia Darwin, Australia
venkatasilpa.borra@cdu.edu.au kamal.debnath@cdu.edu.au

Abstract— This paper presents a comparison between Programming (DP) etc. Nevertheless, heuristic algorithms
Dynamic Programming (DP) and Particle Swarm Optimization such as (1) Tabu search (2) Genetic Algorithm (GA) (3)
(PSO) approaches for minimizing fuel cost and CO2 emissions Particle Swarm Optimization (PSO). These methods are
and solving Unit Commitment (UC) problem in Microgrid suitable for the UC problem. In this paper, DP and PSO
Central Energy Management System (MCEMS). Both techniques are used to solve the UC problem. The MCEMS
approaches minimize the fuel cost and CO2 emissions for the was implemented, which includes PV, battery, MGT and
Micro Gas Turbine (MGT). These techniques are applied to load. In addition, if microgrids can operate in stand-alone
ten subsystems in MCEMS. The MCEMS adjusts itself during mode then it entails a storage system. Microgrids may store
the operation in the generation system. The test results of DP
excess of energy in the battery systems. In this paper, DP and
and PSO are compared with emphasis on a more practical
PSO methods have been compared through applications to
solution. A MATLAB program was written to minimize the
UC problem. Simulation results demonstrate that the PSO
ten subsystems.
technique is more accurate than DP the technique in solving The organization of this paper is as follows: Section II
UC problems. describes the overview of the UC problem. In section III, the
fundamentals of the DP and PSO algorithms are described.
Keywords— Dynamic programming, renewable energy, unit Section IV explains the implementation to study UC. Section
commitment, particle swarm optimization, microgrid.
V mainly focuses on the ten subsystems with loads and
implemented using DP and PSO algorithm. Lastly, results
I. INTRODUCTION are compared with two methods and followed by
Unit commitment (UC) problem is an important and conclusions.
critical problem in power generation. UC refers to the task of
finding optimal scheduling for each generating unit over a II. OVERVIEW OF THE UNIT COMMITMENT
particular time period. It should satisfy the load forecast, PROBLEM
spinning reserve and generation constraints [1-2]. The goal
Researchers have been dealing with UC problem since
of UC is to balance generation and demand while optimizing
the 1940s [6]. Diverse optimization methods are used for
the cost of generation. Electricity markets face many
solving the UC problem.
practical problems in power generation [3]. The problems in
power generation are UC variations due to the change in
demand for electricity due to faults. In addition, 1940s: Appearance 1959s: The first
environmental problems related to the use of fossil fuels, of the UC problem developed to solve the
intermittence of Renewable Energy Sources (RES) and in power system. UC problem.
failure of system components. UC problems may be divided
into three types: traditional UC, Security- Constrained Unit
Commitment (SCUC) and Price-Based Unit Commitment
(PBUC) [4]. UC can be classified by either scheduling in an 1962s: The first 1980s: Critical changes
integrated or deregulated environment. Mixed Integer in power industry
However, the progress in RES has been improved in Programming (MIP) moving to the
recent years to solve environmental problems. It includes formulation for the deregulated
PV, Fuel Cells (FC), Wind Turbines etc. Various computer UC problem. environment.
models have been developed for the design of renewable
power projects. Microgrids contain a cluster of loads and
RES. Such as PV, WT, FC, Micro Gas Turbine (MGT) and
2000s: Integration 2010s: Recent methods
battery storage. Batteries can be used for storage of surplus
of RES in power of UC problem in the
energy. It is mainly utilized for the simulation of grid-
connected or stand-alone mode [5]. system and its presence of intermittent
impact. RES.
Several methods have been proposed to get optimal
generation scheduling. Deterministic algorithms are (1) Fig. 1. Summary of UC problems evolution throughout the years.
Linear Programming (LP) (2) Non-linear Programming
(NLP) (3) Quadratic Programming (QP) (4) Dynamic In Fig. 1, many approaches were used to solve the UC
problem from 1940s to until the present. Nevertheless, The

 
    395
2019 IEEE Jordan International Joint Conference on Electrical Engineering and Information Technology (JEEIT)

UC problem was formulated by considering production cost. B. Particle Swarm Optimization


The total production cost comprises fuel cost and Particle swarm optimization (PSO) is an evolutionary
maintenance cost. It includes the running cost, the operation computation technique. It was originally proposed by
of thermal units, start-up costs, and shutdown costs. Kennedy and Eberhart in the year 1995 [9]. This method is
Considering the start-up/shutdown sequences and the power inspired by bird flocking and fish schooling. It is initialized
outputs for all available units. The UC problem has the with a population of random particles or agents. Usually, it
following form [7]: searches for the optimum solution by updating population
evolution and generations based on the previous
Total production costs = Fuel cost + Start-up cost + generations.
Shutdown cost+ Maintenance cost+ Running cost. Each agent adjusts its speed and direction according to
its own experience. The agents update their velocities and
III. PROPOSED ALGORITHMS positions of each agent. PSO calculates velocity for each
agent based on its previous velocity ( vid ) . Each determines
A. Dynamic Programming
its position by solving for the local solution ( pbestid ) which
The DP schedules the operation of units with
considering lower the fuel cost while satisfying constraints. the best solution (global solution) has been achieved
It controls the operation of the generating units (ON and ( gbest id ) . PSO can be calculated the current velocities and
OFF). Numerous approaches have been proposed to resolve positions of pbest to gbest as shown in the following
efficiently of the UC problem with a better solution. Such equations:
deterministic algorithms are Mixed Integer Programming
(MIP), Genetic Algorithm (GA), Evolutionary Programming
(EP), DP etc. These techniques can handle UC problem. But
vid(t +1) = k * [wvid(t ) + c1r1 ( pbestid − kid(t ) ) + c2 r2 ( gbestid − xid(t ) )] (1)
DP can provide a more accurate solution than other
approaches [8]. This algorithm is a process of solving
problems and finding the local solution one after another. It xid(t+1) = xid(t ) + vid(t+1)
will give several possible solutions for problems, but it (2)
gives only local solution. But DP can solve a problem in its i = 1,2,....,n, d = 1,2....,m
own way. In Fig. 2, the implementation of the DP approach
is given in the flowchart is as follows. wmax − wmin
w = wmax − * iter (3)
itermax
Start
where vid(t ) : the velocity of agent ith in the iteration t ,
Total production cost= Min [fuel d - dimensional space, v d ,min , vid( t ) , v d ,max
cost+ start-up cost+ running cost].
xid(t ) : the current position of agent ith in the iteration t ,

t = t+1 w : inertia weight factor,


t : number of iterations,
Feasible states for n : number of agents in the group,
generating units in interval
k : constriction factor,
t -1.
c1 , c2 : acceleration constants,

Checking for possible r1 , r2 : random numbers between 0 to 1.


solutions for problems. The particle velocity in d- dimensional space is limited with
maximum value v d ,max , N is the total number of particles in

No the swarm. To ensure uniform velocity through all


dimensions, the maximum velocity in a d-dimensional is
Minimizing expressed as follows:
total cost t
=T (Eq.5) vd ,max =
(x id , max − xid ,min )
(4)
N
Yes
PSO gives solutions for problems, but it gives local and
global solution than the DP algorithm. Implementation of the
Stop. An optimal solution for PSO is shown in the flowchart is as follows.
the problem.

Fig. 2. Implementation of the DP flowchart.

396
2019 IEEE Jordan International Joint Conference on Electrical Engineering and Information Technology (JEEIT)

Fi ,t (Pi ,ot ) = a i + bi Pi o,t + c i (Pi ,ot )


2
(6)
Start
The start-up cost is as follows:
­HSCi , T i ,t ≤ MD i ≤ T i ,t + CST i
D ON D
Initialize agents with random velocity SUCi,t = ® (7)
¯CSCi, MD i ≥ T i ,t + CST i
ON D
and position vectors.
where HSCi and CSCi are the hot start-up and cold start-up
Evaluate the fitness of each cost respectively, Ti ,Dt is the time interval of unit i , MDiON is
agent position. the duration during the i th unit is continuously ON and CSTi
is the cold start-up time of unit i .

No V. CASE STUDY
If fitness value The proposed methodology was tested on test identical
is better than subsystems [11]. The proposed approaches have been
fitness (pbest) applied to solve a UC problem, with the given load data
then pbest =p presented in Table II. Each subsystem contains PV, battery
and MGT. Table I contains PV, battery and MGT for ten
Yes subsystems. The output power for every hour is given in
Table II.
Set best of pbests and gbest. TABLE I. OUTPUT DATA
Subsystem PV (kW) Battery (kW) MGT (kW)
1 400 400 55
2 400 400 55
No 3 100 100 30
Update agent 4 100 100 30
velocity (Eq. 1) 5 150 150 15
and position 6 50 50 30
7 50 50 35
(Eq.2) 8 35 35 15
9 35 35 15
Yes 10 15 15 15

TABLE II. LOAD DATA


Stop. Optimal solution is gbest.
Demand Demand
Fig. 3. Implementation of the PSO flowchart. Hour (kW) Hour (kW)
1 700 13 1400
2 750 14 1300
IV. IMPLEMENTATION TO SOLVE UNIT 3 850 15 1200
COMMITMENT PROBLEM 4 950 16 1050
5 1000 17 1000
The fuel cost of the generating units should be considered
6 1100 18 1100
in system operation as an objective function of a UC 7 1150 19 1200
problem. It includes the start-up costs of the committed units, 8 1200 20 1400
the fuel costs of the generating units and the shut-down costs 9 1300 21 1300
of the de-committed units. The function can be expressed by 10 1400 22 1100
the below equation. 11 1450 23 900
12 1500 24 800
­T N T N
Minimize ®¦¦Fi,t (Pio,t ) *ui,t + ¦¦SUCi,t *ui,t
¯t=1 i=1 t=1 i=1 VI. RESULTS AND DISCUSSION
T N A test scenario is conducted by using DP and PSO. In
(1 − ui,t−1 ) + ¦¦SDCi,t * ui,t −1 (1 − ui,t )} (5) order to compare the effectiveness of the two methods. DP
t =1 i =1 and PSO are applied independently in subsystems
considering fuel cost and emissions.
where Pi o,t is the power output at a time t of unit i , ui ,t is
The DP was tested on ten subsystems each containing its
the ON and OFF status of unit i at time t , SUCi ,t and SDCi ,t own power source. It is using PV generation, battery and
are the start-up and shutdown cost of unit i at time t , N is MGT to generate electricity, according to demand. In Fig. 4,
the number of units, T is the particular time interval and Fi ,t subsystems are generating power using PV, battery and
MGT. The 1st and 2nd subsystems are generating 400 kW
is the fuel cost. The fuel costs of the generating units are from PV and 55 kW from MGT respectively. The 4th and 5th
normally given as a quadratic form [10] in Eq. (6). subsystems are generating 130 kW and 60 kW respectively.

397
2019 IEEE Jordan International Joint Conference on Electrical Engineering and Information Technology (JEEIT)

90 kW. Instead of using 3rd and 4th MGT generation in


Fig. 6, PSO is using available PV generation in the 5th
subsystem.

Fig. 4. Subsystems 1, 2, 4 and 5 are committed in the DP approach.

On the other hand, PSO is tested on ten subsystems by Fig. 6. Subsystems 1, 2, 3, 4 and 5 are committed in the DP approach.
one-hour intervals for 24 hours. In Fig. 5, 1st, 2nd, 3rd, 4th and
5th subsystems are generating power from PV in the 6th hour.
By comparing Fig 4 and 5, PSO is successfully using PV
than MGT, which minimizes the fuel cost.

Fig. 7. Subsystems 1, 2, 3, 4 and 5 are committed in the PSO approach.

In Table III denotes the DP performance results. It lists


Fig. 5. Subsystems 1, 2, 3, 4 and 5 are committed in the PSO approach. information about the PV, MGT generation and battery along
with subsystems. These results were obtained for solving the
In the DP, subsystems 1 and 2 are generating power 300 UC problem by using DP.
kW from PV, 55 kW from MGT and 100 kW from the
battery in Fig. 6. In addition, 3rd and 4th subsystems are TABLE III DP PERFORMANCE RESULTS
generating 90 kW from PV, 30 kW from MGT and 10 kW Hour Subsystem 1 Subsystem 2 Subsystem 5
from the battery. The 5th subsystem is generating 30 kW
PV MGT PV MGT PV Bat
from PV. MGTs are generating power in subsystems 1, 2, 3
and 4 to satisfy the demand. 1 400 55 245 0 0 0
st nd 2 400 55 295 0 0 0
In Fig. 7, 1 and 2 subsystems are generating power
395 kW from PV, 55 kW from MGT and 5 kW from the 3 400 50 400 0 0 0
battery by using the PSO approach. In addition, 3rd, 4th and
5th subsystems are generating 100 kW and 90 kW from PV. 4 400 55 400 5 40 5
The 5th subsystem is increased generation from 30 kW to 5 400 55 400 55 85 5

398
2019 IEEE Jordan International Joint Conference on Electrical Engineering and Information Technology (JEEIT)

In Table III subsystems 1, 2 and 5 are mentioned for five MGT generation for 24 hours. The average cost of MGT
hours. In Table IV demonstrates the PSO performance generation per unit in Australia is $0.024. It has been
results. This Table denotes PV, MGT generation and battery considered fuel cost.
with subsystems using PSO. The trend of finding the best
solution is revealed in the below Table.

TABLE IV PSO PERFORMANCE RESULTS


Hour Subsystem 2 3 4 5
1
PV Bat PV Bat PV PV PV

1 395 5 295 5 0 0 0
2 395 5 345 5 0 0 0
3 395 5 395 5 0 0 0
4 395 5 395 5 0 100 55
5 400 0 400 0 50 100 150

When PV and battery are insufficient to generate the


required amount of power then an MGT will be used. But,
DP is using MGT than PV generation in some cases.
Because of DP is solving the problem in its own way. It also
provides a local solution with considering immediate
neighbours. Because of these drawbacks, PSO is Fig. 9. Cost of MGT generation.
implemented in the next scenario. alternate generation. In the
PSO, the first decision is to use PV and battery before MGT The emissions are minimized in the PSO technique in
is used because of PV generation is cheaper and pollution Fig. 10. Because PSO technique is using to generate
free, unlike MGTs. By comparing with DP, PSO can be used electricity from PV generation. The PSO approach
to generate PV generation. PSO provides a global solution. minimizes the UC problem, fuel cost and emissions rather
than DP.
The PV generation cost includes losses, labour charges,
bank interest rates and maintenance etc. In Fig. 8, the graph
gives information about the cost analysis of PV generation
for every 24 hours. The average cost of PV generation per
unit in Australia is $0.016. The costs of generation are shown
in Fig. 8. PSO is using more PV generation than DP.

Fig. 10. CO2 emissions are comparing with DP and PSO approaches.

VII. CONCLUSION

The Particle Swarm Optimization (PSO) and Dynamic


Fig. 8. Cost of PV generation. Programming (DP) techniques were successfully applied for
By observing Fig. 9, DP was using MGT generation than ten subsystems. The results are compared to check the
PV generation. The power generated from PV and MGT is effectiveness of the methods adopted from the literature.
separated for each hour. This Figure describes the cost of Due to the limitations of local solutions in DP, it may

399
2019 IEEE Jordan International Joint Conference on Electrical Engineering and Information Technology (JEEIT)

dictate the use of Micro Gas Turbine (MGT) incorrectly, Technology and Advanced Engineering, vol. 4, Issue 4, pp. 880-884,
2014.
whereas PV generation is still available. On the other hand,
[6] R. B. Johnson, Chao-An Li and A. J.Svoboda, “ A new Unit
the PSO approach recommend the use of PV generation as Commitment Method,” IEEE Trans. On Power Systems, vol. 12,
long as this is available. This technique also used MGT Issue 1, pp. 113-119, Feb. 1997.
where PV generation and battery are not available. The [7] J. P. S. Catalao, “Smart and Sustainable Power Systems: Operations,
obtained results show that the PSO is effectively solving Planning and Economics of Insular Electricity Grids,” CRC Press,
Unit Commitment problem by considering cost of June 2015.
generation and CO2 emissions rather than DP. [8] R. Naresh Sharma and Prateek Kumar, “Dynamic Programming
Approach for Solving Power Generating Unit Commitment Problem,”
International Conference on Computer and Communication
Technology, pp. 298-303, Sep. 2011.
ACKNOWLEDGEMENT [9] J. Kennedy, R. Eberhart, "Particle Swarm Optimization," in
The authors would like to acknowledge the financial Proceedings of IEEE International Conference on Neural Networks,
support provided by Charles Darwin University and the vol. 4, pp. 1942-1948, Nov. 1995.
Australian Commonwealth Government. [10] S. Jadid, M. P. Moghaddam and A. Badri, “The impact of generators
on Equilibrium Considering Transmission Constraints,” European
Trans. On Electrical Power, vol. 19, pp. 765-777, 2009.
[11] V. S. Pappala and I. Erlich, “A new approach for solving the UC
REFERENCES problem by adaptive particle swarm optimization,” in Proc. IEEE, pp.
1-6, 2008.

[1] Mohamed Tkiouat and I. Abdou, “Unit Commitment Problem in


Electrical Power System: A Literature Review,” International Journal Venkata Silpa Borra received her Bachelor’s degree in
of Electrical and Computer Engineering, vol. 8, No. 3, pp. 1357-1372, electrical and electronics engineering and Master’s degree in
June 2018. power system control and automation from the Jawaharlal
[2] V. Miranda, O. R. Saavedra and Mauro S.S. Pinto, “Risk and Unit Nehru Technological University, Kakinada, India. She is
Commitment Decisions in Scenarios of Renewable Energy
Uncertainty,” Elsevier, vol. 97, pp. 550-558. currently a Master by Research student at the Charles
[3] J. J. Jamian, S. Y. Abujarad and M. W. Mustafa, “Recent Approaches Darwin University, Australia. She is a graduate student
of Unit Commitment in the Presence of Intermittent Renewable member of IEEE and Engineers Australia.
Energy Resources,” Econpapers, vol. 70, pp. 215-223, 2017.
[4] H. Y. Yamin, “Review on Methods of Generation Scheduling in K. Debnath taught power systems in various Universities in
Electric Power Systems,” Electric Power Systems Research, vol. 69, Australia as well as one in Singapore. Currently, he is a staff
no. 4, pp. 227-248, 2004.
[5] Arun Shandilya and Anil Gupta, “Challenges of Integration of Wind
member of Charles Darwin University, Australia. Dr.
Power on Power System Grid,” International Journal of Emerging Debnath is a member of IEEE and Engineers Australia.

400

You might also like